Fuzzy ARTMAP with Feature Weighting
Răzvan Andonie, Angel Caţaron, Lucian Mircea Sasu · International conference on Artificial intelligence and applications · 2008
We introduce a novel Fuzzy ARTMAP (FAM) architecture: FAM with Feature Weighting (FAMFW). In the first stage, the features of the training data are weighted. In the second stage, the obtained weights are used to improve the FAMFW training. The effect of this approach is a more sensitive FAM category determination: Category dimensions in the direction of relevant features are decreased whereas category dimensions in the direction of non-relevant feature are increased. Potentially, any feature weighting method could be used, which makes the FAMFW very general. In our study, we use a feature weighting algorithm based on the Neural-Gas algorithm.